ServiceX REST Server

https://travis-ci.org/ssl-hep/ServiceX_transformer.svg?branch=pytest

This is a flask app for receiving user requests and orchestrating the ServiceX services:

User Management

If ENABLE_AUTH is set to True, an identity management system will be enabled. Users will need to create accounts in order to make requests. The system consists of two components: authentication (verification of each user's identity) and authorization (control of access to API resources).

Authentication

Authentication is currently implemented via Globus, a federated identity provider which implements the OAuth 2.0 protocol. Prospective users must visit the ServiceX website included in this Flask app at its hosted domain. There, they can sign in via Globus to confirm their identity using any supported identity provider (e.g. their university, CILogon, or GitHub account). A corresponding ServiceX user account will be created. Existing users can also visit the website to view information associated with their account.

Authorization

If authentication is enabled, API resources will be protected with JWT bearer tokens. Authenticated users will be issued a ServiceX API token, which is a JWT refresh token containing their identity. This refresh token can be used to obtain access tokens, which are required in order to access protected resources.

Clients accessing the ServiceX API, such as the Python client, should accept the API token as a configuration value or environment variable, and use it to obtain access tokens and make requests. Requests to secured endpoints should have an HTTP header called Authorization with a value of Bearer <access_token>.

Administrators

Admin accounts are used to manage other users. The API server supports a JWT_ADMIN config property, which should be set to the email address of the initial admin user. The first user to sign up with this email address will be made an admin user automatically. This should be done immediately after deployment.

New users are marked as pending, and will be unable to submit requests until approved. ServiceX admins can view the pending users with a GET on the /pending endpoint. They can approve a pending user with a POST to /accept with a body of:

{
    "email": "<the email>"
}

The API can also be configured to send notifications of new user registrations to a Slack channel of choice, allowing administrators to approve pending users directly from Slack. This requires setting up a Slack app and supplying the SLACK_SIGNING_SECRET and SIGNUP_WEBHOOK_URL config values. For full details, see the ServiceX README.

Email notifications can be sent to users when they are approved via Mailgun. All this requires is setting the MAILGUN_API_KEY and MAILGUN_DOMAIN config values.

Local Development

First install the app and test dependencies with:

pip install -e ".[test]"

The application depends on the psycopg2 library to access postgres. The install for this library can be tricky on some desktop environments. You may have better luck with the pre-built binary package. Edit setup.py and replace psycopg2 with psycopg2-binary and try to pip install again.

Running Tests

We use flake8 to verify coding standards. You can run this tool over your code with the command

flake8

We use pytest to verify the code. These tests can be run with the command

python -m pytest

Detailed Transformer Logging

It can be complicated to debug issues inside the transformers on a very large dataset. To make it easier, detailed log statements are sent back from the transformers and stored in the postgres database. If you need to interrogate this database you can forward the postgres port to your development system with

kubectl port-forward xaod-postgresql-0 5432:5432

Then use your favorite postgres sql client to connect to this database with the connection URL``jdbc:postgresql://localhost:5432/postgres`` with the user postgres and the password leftfoot1.

Of particular interest is the file_status table. It contains entries for each transformed root file when it starts and when it finishes, retries, or reports an error.

You can find errors with

select * from file_status where request_id='da3d7cc2-7f97-4c98-be77-4e9c22f67425' and status='failure';

These records include the error and the first 2048 characters of the stack trace along with the pod name where the transformer ran (you might be able to catch the full log with the kubectl logs -p which can retreive logs from terminated pods)

More information on the specific file can be found by joining with the files table.

Database Migration

We use the alembic tooling under Flask-Migrate library to manage migration of database as schema evolves.

Alembic stores database versions in the migrations directory. These are applied to the database as part of boot.sh

Create a new version of the database after updating models.py with the command:

FLASK_APP=servicex/app.py APP_CONFIG_FILE=../app.conf flask db migrate -m "<<release version>>"

This will add a new script under migrations directory. Check it for accuracy and check it into the repo to have it applied when the built docker image is next run against a database.

Building Docker Image

docker build -t servicex_app .

Running Docker

docker run --name servicex-app --rm -p8000:5000 \
 --mount type=bind,source="$(pwd)"/sqlite,target=/sqlite \
 -e APP_CONFIG_FILE=/home/servicex/docker-dev.conf \
 servicex_app:latest

Cleaning up old Transformation Queues

It's easy to accumulate a bunch of transformation queues during testing. It would be quite tedious to try to delete them via the management console. You can install the rabbitmqadmin cli and then with some tricky scripting batch delete queues:

./d.sh $(python  rabbitmqadmin -V / --port=30182 -u user -p leftfoot1 list queues | grep ".*-.*" | awk '{print $2}')

ssl-hep/ServiceX_App